fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "xattr.h"
  64 #include "acl.h"
  65 #include "ocfs2_trace.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup,
  83                                     bool dio);
  84
  85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  86                             handle_t *handle,
  87                             struct inode *inode,
  88                             struct buffer_head *fe_bh,
  89                             char *name,
  90                             struct ocfs2_dir_lookup_result *lookup,
  91                             struct inode *orphan_dir_inode,
  92                             bool dio);
  93
  94 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  95                                      handle_t *handle,
  96                                      struct inode *inode,
  97                                      const char *symname);
  98
  99 static int ocfs2_double_lock(struct ocfs2_super *osb,
 100                              struct buffer_head **bh1,
 101                              struct inode *inode1,
 102                              struct buffer_head **bh2,
 103                              struct inode *inode2,
 104                              int rename);
 105
 106 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
 107 /* An orphan dir name is an 8 byte value, printed as a hex string */
 108 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 109
 110 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 111                                    unsigned int flags)
 112 {
 113         int status;
 114         u64 blkno;
 115         struct inode *inode = NULL;
 116         struct dentry *ret;
 117         struct ocfs2_inode_info *oi;
 118
 119         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 120                            dentry->d_name.name,
 121                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 122
 123         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 124                 ret = ERR_PTR(-ENAMETOOLONG);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 129         if (status < 0) {
 130                 if (status != -ENOENT)
 131                         mlog_errno(status);
 132                 ret = ERR_PTR(status);
 133                 goto bail;
 134         }
 135
 136         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 137                                             dentry->d_name.len, &blkno);
 138         if (status < 0)
 139                 goto bail_add;
 140
 141         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 142         if (IS_ERR(inode)) {
 143                 ret = ERR_PTR(-EACCES);
 144                 goto bail_unlock;
 145         }
 146
 147         oi = OCFS2_I(inode);
 148         /* Clear any orphaned state... If we were able to look up the
 149          * inode from a directory, it certainly can't be orphaned. We
 150          * might have the bad state from a node which intended to
 151          * orphan this inode but crashed before it could commit the
 152          * unlink. */
 153         spin_lock(&oi->ip_lock);
 154         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 155         spin_unlock(&oi->ip_lock);
 156
 157 bail_add:
 158         ret = d_splice_alias(inode, dentry);
 159
 160         if (inode) {
 161                 /*
 162                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 163                  * dentry, it will d_move() it on top of ourse. The
 164                  * return value will indicate this however, so in
 165                  * those cases, we switch them around for the locking
 166                  * code.
 167                  *
 168                  * NOTE: This dentry already has ->d_op set from
 169                  * ocfs2_get_parent() and ocfs2_get_dentry()
 170                  */
 171                 if (!IS_ERR_OR_NULL(ret))
 172                         dentry = ret;
 173
 174                 status = ocfs2_dentry_attach_lock(dentry, inode,
 175                                                   OCFS2_I(dir)->ip_blkno);
 176                 if (status) {
 177                         mlog_errno(status);
 178                         ret = ERR_PTR(status);
 179                         goto bail_unlock;
 180                 }
 181         } else
 182                 ocfs2_dentry_attach_gen(dentry);
 183
 184 bail_unlock:
 185         /* Don't drop the cluster lock until *after* the d_add --
 186          * unlink on another node will message us to remove that
 187          * dentry under this lock so otherwise we can race this with
 188          * the downconvert thread and have a stale dentry. */
 189         ocfs2_inode_unlock(dir, 0);
 190
 191 bail:
 192
 193         trace_ocfs2_lookup_ret(ret);
 194
 195         return ret;
 196 }
 197
 198 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 199 {
 200         struct inode *inode;
 201         int status;
 202
 203         inode = new_inode(dir->i_sb);
 204         if (!inode) {
 205                 mlog(ML_ERROR, "new_inode failed!\n");
 206                 return ERR_PTR(-ENOMEM);
 207         }
 208
 209         /* populate as many fields early on as possible - many of
 210          * these are used by the support functions here and in
 211          * callers. */
 212         if (S_ISDIR(mode))
 213                 set_nlink(inode, 2);
 214         inode_init_owner(inode, dir, mode);
 215         status = dquot_initialize(inode);
 216         if (status)
 217                 return ERR_PTR(status);
 218
 219         return inode;
 220 }
 221
 222 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 223                 struct dentry *dentry, struct inode *inode)
 224 {
 225         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 226
 227         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 228         ocfs2_lock_res_free(&dl->dl_lockres);
 229         BUG_ON(dl->dl_count != 1);
 230         spin_lock(&dentry_attach_lock);
 231         dentry->d_fsdata = NULL;
 232         spin_unlock(&dentry_attach_lock);
 233         kfree(dl);
 234         iput(inode);
 235 }
 236
 237 static int ocfs2_mknod(struct inode *dir,
 238                        struct dentry *dentry,
 239                        umode_t mode,
 240                        dev_t dev)
 241 {
 242         int status = 0;
 243         struct buffer_head *parent_fe_bh = NULL;
 244         handle_t *handle = NULL;
 245         struct ocfs2_super *osb;
 246         struct ocfs2_dinode *dirfe;
 247         struct buffer_head *new_fe_bh = NULL;
 248         struct inode *inode = NULL;
 249         struct ocfs2_alloc_context *inode_ac = NULL;
 250         struct ocfs2_alloc_context *data_ac = NULL;
 251         struct ocfs2_alloc_context *meta_ac = NULL;
 252         int want_clusters = 0;
 253         int want_meta = 0;
 254         int xattr_credits = 0;
 255         struct ocfs2_security_xattr_info si = {
 256                 .enable = 1,
 257         };
 258         int did_quota_inode = 0;
 259         struct ocfs2_dir_lookup_result lookup = { NULL, };
 260         sigset_t oldset;
 261         int did_block_signals = 0;
 262         struct posix_acl *default_acl = NULL, *acl = NULL;
 263         struct ocfs2_dentry_lock *dl = NULL;
 264
 265         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 266                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 267                           (unsigned long)dev, mode);
 268
 269         status = dquot_initialize(dir);
 270         if (status) {
 271                 mlog_errno(status);
 272                 return status;
 273         }
 274
 275         /* get our super block */
 276         osb = OCFS2_SB(dir->i_sb);
 277
 278         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 279         if (status < 0) {
 280                 if (status != -ENOENT)
 281                         mlog_errno(status);
 282                 return status;
 283         }
 284
 285         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 286                 status = -EMLINK;
 287                 goto leave;
 288         }
 289
 290         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 291         if (!ocfs2_read_links_count(dirfe)) {
 292                 /* can't make a file in a deleted directory. */
 293                 status = -ENOENT;
 294                 goto leave;
 295         }
 296
 297         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 298                                            dentry->d_name.len);
 299         if (status)
 300                 goto leave;
 301
 302         /* get a spot inside the dir. */
 303         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 304                                               dentry->d_name.name,
 305                                               dentry->d_name.len, &lookup);
 306         if (status < 0) {
 307                 mlog_errno(status);
 308                 goto leave;
 309         }
 310
 311         /* reserve an inode spot */
 312         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 313         if (status < 0) {
 314                 if (status != -ENOSPC)
 315                         mlog_errno(status);
 316                 goto leave;
 317         }
 318
 319         inode = ocfs2_get_init_inode(dir, mode);
 320         if (IS_ERR(inode)) {
 321                 status = PTR_ERR(inode);
 322                 inode = NULL;
 323                 mlog_errno(status);
 324                 goto leave;
 325         }
 326
 327         /* get security xattr */
 328         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 329         if (status) {
 330                 if (status == -EOPNOTSUPP)
 331                         si.enable = 0;
 332                 else {
 333                         mlog_errno(status);
 334                         goto leave;
 335                 }
 336         }
 337
 338         /* calculate meta data/clusters for setting security and acl xattr */
 339         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 340                                        &si, &want_clusters,
 341                                        &xattr_credits, &want_meta);
 342         if (status < 0) {
 343                 mlog_errno(status);
 344                 goto leave;
 345         }
 346
 347         /* Reserve a cluster if creating an extent based directory. */
 348         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 349                 want_clusters += 1;
 350
 351                 /* Dir indexing requires extra space as well */
 352                 if (ocfs2_supports_indexed_dirs(osb))
 353                         want_meta++;
 354         }
 355
 356         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 357         if (status < 0) {
 358                 if (status != -ENOSPC)
 359                         mlog_errno(status);
 360                 goto leave;
 361         }
 362
 363         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 364         if (status < 0) {
 365                 if (status != -ENOSPC)
 366                         mlog_errno(status);
 367                 goto leave;
 368         }
 369
 370         status = posix_acl_create(dir, &mode, &default_acl, &acl);
 371         if (status) {
 372                 mlog_errno(status);
 373                 goto leave;
 374         }
 375         /* update inode->i_mode after mask with "umask". */
 376         inode->i_mode = mode;
 377
 378         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 379                                                             S_ISDIR(mode),
 380                                                             xattr_credits));
 381         if (IS_ERR(handle)) {
 382                 status = PTR_ERR(handle);
 383                 handle = NULL;
 384                 mlog_errno(status);
 385                 goto leave;
 386         }
 387
 388         /* Starting to change things, restart is no longer possible. */
 389         ocfs2_block_signals(&oldset);
 390         did_block_signals = 1;
 391
 392         status = dquot_alloc_inode(inode);
 393         if (status)
 394                 goto leave;
 395         did_quota_inode = 1;
 396
 397         /* do the real work now. */
 398         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 399                                     &new_fe_bh, parent_fe_bh, handle,
 400                                     inode_ac);
 401         if (status < 0) {
 402                 mlog_errno(status);
 403                 goto leave;
 404         }
 405
 406         if (S_ISDIR(mode)) {
 407                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 408                                             new_fe_bh, data_ac, meta_ac);
 409                 if (status < 0) {
 410                         mlog_errno(status);
 411                         goto leave;
 412                 }
 413
 414                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 415                                                  parent_fe_bh,
 416                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 417                 if (status < 0) {
 418                         mlog_errno(status);
 419                         goto leave;
 420                 }
 421                 ocfs2_add_links_count(dirfe, 1);
 422                 ocfs2_journal_dirty(handle, parent_fe_bh);
 423                 inc_nlink(dir);
 424         }
 425
 426         if (default_acl) {
 427                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 428                                        ACL_TYPE_DEFAULT, default_acl,
 429                                        meta_ac, data_ac);
 430         }
 431         if (!status && acl) {
 432                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 433                                        ACL_TYPE_ACCESS, acl,
 434                                        meta_ac, data_ac);
 435         }
 436
 437         if (status < 0) {
 438                 mlog_errno(status);
 439                 goto leave;
 440         }
 441
 442         if (si.enable) {
 443                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 444                                                  meta_ac, data_ac);
 445                 if (status < 0) {
 446                         mlog_errno(status);
 447                         goto leave;
 448                 }
 449         }
 450
 451         /*
 452          * Do this before adding the entry to the directory. We add
 453          * also set d_op after success so that ->d_iput() will cleanup
 454          * the dentry lock even if ocfs2_add_entry() fails below.
 455          */
 456         status = ocfs2_dentry_attach_lock(dentry, inode,
 457                                           OCFS2_I(dir)->ip_blkno);
 458         if (status) {
 459                 mlog_errno(status);
 460                 goto leave;
 461         }
 462
 463         dl = dentry->d_fsdata;
 464
 465         status = ocfs2_add_entry(handle, dentry, inode,
 466                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 467                                  &lookup);
 468         if (status < 0) {
 469                 mlog_errno(status);
 470                 goto leave;
 471         }
 472
 473         insert_inode_hash(inode);
 474         d_instantiate(dentry, inode);
 475         status = 0;
 476 leave:
 477         if (default_acl)
 478                 posix_acl_release(default_acl);
 479         if (acl)
 480                 posix_acl_release(acl);
 481         if (status < 0 && did_quota_inode)
 482                 dquot_free_inode(inode);
 483         if (handle)
 484                 ocfs2_commit_trans(osb, handle);
 485
 486         ocfs2_inode_unlock(dir, 1);
 487         if (did_block_signals)
 488                 ocfs2_unblock_signals(&oldset);
 489
 490         brelse(new_fe_bh);
 491         brelse(parent_fe_bh);
 492         kfree(si.value);
 493
 494         ocfs2_free_dir_lookup_result(&lookup);
 495
 496         if (inode_ac)
 497                 ocfs2_free_alloc_context(inode_ac);
 498
 499         if (data_ac)
 500                 ocfs2_free_alloc_context(data_ac);
 501
 502         if (meta_ac)
 503                 ocfs2_free_alloc_context(meta_ac);
 504
 505         /*
 506          * We should call iput after the i_mutex of the bitmap been
 507          * unlocked in ocfs2_free_alloc_context, or the
 508          * ocfs2_delete_inode will mutex_lock again.
 509          */
 510         if ((status < 0) && inode) {
 511                 if (dl)
 512                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 513
 514                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 515                 clear_nlink(inode);
 516                 iput(inode);
 517         }
 518
 519         if (status)
 520                 mlog_errno(status);
 521
 522         return status;
 523 }
 524
 525 static int __ocfs2_mknod_locked(struct inode *dir,
 526                                 struct inode *inode,
 527                                 dev_t dev,
 528                                 struct buffer_head **new_fe_bh,
 529                                 struct buffer_head *parent_fe_bh,
 530                                 handle_t *handle,
 531                                 struct ocfs2_alloc_context *inode_ac,
 532                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 533 {
 534         int status = 0;
 535         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 536         struct ocfs2_dinode *fe = NULL;
 537         struct ocfs2_extent_list *fel;
 538         u16 feat;
 539         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 540
 541         *new_fe_bh = NULL;
 542
 543         /* populate as many fields early on as possible - many of
 544          * these are used by the support functions here and in
 545          * callers. */
 546         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 547         OCFS2_I(inode)->ip_blkno = fe_blkno;
 548         spin_lock(&osb->osb_lock);
 549         inode->i_generation = osb->s_next_generation++;
 550         spin_unlock(&osb->osb_lock);
 551
 552         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 553         if (!*new_fe_bh) {
 554                 status = -ENOMEM;
 555                 mlog_errno(status);
 556                 goto leave;
 557         }
 558         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 559
 560         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 561                                          *new_fe_bh,
 562                                          OCFS2_JOURNAL_ACCESS_CREATE);
 563         if (status < 0) {
 564                 mlog_errno(status);
 565                 goto leave;
 566         }
 567
 568         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 569         memset(fe, 0, osb->sb->s_blocksize);
 570
 571         fe->i_generation = cpu_to_le32(inode->i_generation);
 572         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 573         fe->i_blkno = cpu_to_le64(fe_blkno);
 574         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 575         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 576         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 577         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 578         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 579         fe->i_mode = cpu_to_le16(inode->i_mode);
 580         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 581                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 582
 583         ocfs2_set_links_count(fe, inode->i_nlink);
 584
 585         fe->i_last_eb_blk = 0;
 586         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 587         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 588         fe->i_atime = fe->i_ctime = fe->i_mtime =
 589                 cpu_to_le64(CURRENT_TIME.tv_sec);
 590         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 591                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 592         fe->i_dtime = 0;
 593
 594         /*
 595          * If supported, directories start with inline data. If inline
 596          * isn't supported, but indexing is, we start them as indexed.
 597          */
 598         feat = le16_to_cpu(fe->i_dyn_features);
 599         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 600                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 601
 602                 fe->id2.i_data.id_count = cpu_to_le16(
 603                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 604         } else {
 605                 fel = &fe->id2.i_list;
 606                 fel->l_tree_depth = 0;
 607                 fel->l_next_free_rec = 0;
 608                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 609         }
 610
 611         ocfs2_journal_dirty(handle, *new_fe_bh);
 612
 613         ocfs2_populate_inode(inode, fe, 1);
 614         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 615         if (!ocfs2_mount_local(osb)) {
 616                 status = ocfs2_create_new_inode_locks(inode);
 617                 if (status < 0)
 618                         mlog_errno(status);
 619         }
 620
 621         oi->i_sync_tid = handle->h_transaction->t_tid;
 622         oi->i_datasync_tid = handle->h_transaction->t_tid;
 623
 624 leave:
 625         if (status < 0) {
 626                 if (*new_fe_bh) {
 627                         brelse(*new_fe_bh);
 628                         *new_fe_bh = NULL;
 629                 }
 630         }
 631
 632         if (status)
 633                 mlog_errno(status);
 634         return status;
 635 }
 636
 637 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 638                               struct inode *dir,
 639                               struct inode *inode,
 640                               dev_t dev,
 641                               struct buffer_head **new_fe_bh,
 642                               struct buffer_head *parent_fe_bh,
 643                               handle_t *handle,
 644                               struct ocfs2_alloc_context *inode_ac)
 645 {
 646         int status = 0;
 647         u64 suballoc_loc, fe_blkno = 0;
 648         u16 suballoc_bit;
 649
 650         *new_fe_bh = NULL;
 651
 652         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 653                                        inode_ac, &suballoc_loc,
 654                                        &suballoc_bit, &fe_blkno);
 655         if (status < 0) {
 656                 mlog_errno(status);
 657                 return status;
 658         }
 659
 660         status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 661                                     parent_fe_bh, handle, inode_ac,
 662                                     fe_blkno, suballoc_loc, suballoc_bit);
 663         if (status < 0) {
 664                 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
 665                 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
 666                                 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
 667                 if (tmp)
 668                         mlog_errno(tmp);
 669         }
 670
 671         return status;
 672 }
 673
 674 static int ocfs2_mkdir(struct inode *dir,
 675                        struct dentry *dentry,
 676                        umode_t mode)
 677 {
 678         int ret;
 679
 680         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 681                           OCFS2_I(dir)->ip_blkno, mode);
 682         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 683         if (ret)
 684                 mlog_errno(ret);
 685
 686         return ret;
 687 }
 688
 689 static int ocfs2_create(struct inode *dir,
 690                         struct dentry *dentry,
 691                         umode_t mode,
 692                         bool excl)
 693 {
 694         int ret;
 695
 696         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 697                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 698         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 699         if (ret)
 700                 mlog_errno(ret);
 701
 702         return ret;
 703 }
 704
 705 static int ocfs2_link(struct dentry *old_dentry,
 706                       struct inode *dir,
 707                       struct dentry *dentry)
 708 {
 709         handle_t *handle;
 710         struct inode *inode = d_inode(old_dentry);
 711         struct inode *old_dir = d_inode(old_dentry->d_parent);
 712         int err;
 713         struct buffer_head *fe_bh = NULL;
 714         struct buffer_head *old_dir_bh = NULL;
 715         struct buffer_head *parent_fe_bh = NULL;
 716         struct ocfs2_dinode *fe = NULL;
 717         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 718         struct ocfs2_dir_lookup_result lookup = { NULL, };
 719         sigset_t oldset;
 720         u64 old_de_ino;
 721
 722         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 723                          old_dentry->d_name.len, old_dentry->d_name.name,
 724                          dentry->d_name.len, dentry->d_name.name);
 725
 726         if (S_ISDIR(inode->i_mode))
 727                 return -EPERM;
 728
 729         err = dquot_initialize(dir);
 730         if (err) {
 731                 mlog_errno(err);
 732                 return err;
 733         }
 734
 735         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 736                         &parent_fe_bh, dir, 0);
 737         if (err < 0) {
 738                 if (err != -ENOENT)
 739                         mlog_errno(err);
 740                 return err;
 741         }
 742
 743         /* make sure both dirs have bhs
 744          * get an extra ref on old_dir_bh if old==new */
 745         if (!parent_fe_bh) {
 746                 if (old_dir_bh) {
 747                         parent_fe_bh = old_dir_bh;
 748                         get_bh(parent_fe_bh);
 749                 } else {
 750                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 751                         err = -EIO;
 752                         goto out;
 753                 }
 754         }
 755
 756         if (!dir->i_nlink) {
 757                 err = -ENOENT;
 758                 goto out;
 759         }
 760
 761         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 762                         old_dentry->d_name.len, &old_de_ino);
 763         if (err) {
 764                 err = -ENOENT;
 765                 goto out;
 766         }
 767
 768         /*
 769          * Check whether another node removed the source inode while we
 770          * were in the vfs.
 771          */
 772         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 773                 err = -ENOENT;
 774                 goto out;
 775         }
 776
 777         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 778                                         dentry->d_name.len);
 779         if (err)
 780                 goto out;
 781
 782         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 783                                            dentry->d_name.name,
 784                                            dentry->d_name.len, &lookup);
 785         if (err < 0) {
 786                 mlog_errno(err);
 787                 goto out;
 788         }
 789
 790         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 791         if (err < 0) {
 792                 if (err != -ENOENT)
 793                         mlog_errno(err);
 794                 goto out;
 795         }
 796
 797         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 798         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 799                 err = -EMLINK;
 800                 goto out_unlock_inode;
 801         }
 802
 803         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 804         if (IS_ERR(handle)) {
 805                 err = PTR_ERR(handle);
 806                 handle = NULL;
 807                 mlog_errno(err);
 808                 goto out_unlock_inode;
 809         }
 810
 811         /* Starting to change things, restart is no longer possible. */
 812         ocfs2_block_signals(&oldset);
 813
 814         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 815                                       OCFS2_JOURNAL_ACCESS_WRITE);
 816         if (err < 0) {
 817                 mlog_errno(err);
 818                 goto out_commit;
 819         }
 820
 821         inc_nlink(inode);
 822         inode->i_ctime = CURRENT_TIME;
 823         ocfs2_set_links_count(fe, inode->i_nlink);
 824         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 825         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 826         ocfs2_journal_dirty(handle, fe_bh);
 827
 828         err = ocfs2_add_entry(handle, dentry, inode,
 829                               OCFS2_I(inode)->ip_blkno,
 830                               parent_fe_bh, &lookup);
 831         if (err) {
 832                 ocfs2_add_links_count(fe, -1);
 833                 drop_nlink(inode);
 834                 mlog_errno(err);
 835                 goto out_commit;
 836         }
 837
 838         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 839         if (err) {
 840                 mlog_errno(err);
 841                 goto out_commit;
 842         }
 843
 844         ihold(inode);
 845         d_instantiate(dentry, inode);
 846
 847 out_commit:
 848         ocfs2_commit_trans(osb, handle);
 849         ocfs2_unblock_signals(&oldset);
 850 out_unlock_inode:
 851         ocfs2_inode_unlock(inode, 1);
 852
 853 out:
 854         ocfs2_double_unlock(old_dir, dir);
 855
 856         brelse(fe_bh);
 857         brelse(parent_fe_bh);
 858         brelse(old_dir_bh);
 859
 860         ocfs2_free_dir_lookup_result(&lookup);
 861
 862         if (err)
 863                 mlog_errno(err);
 864
 865         return err;
 866 }
 867
 868 /*
 869  * Takes and drops an exclusive lock on the given dentry. This will
 870  * force other nodes to drop it.
 871  */
 872 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 873 {
 874         int ret;
 875
 876         ret = ocfs2_dentry_lock(dentry, 1);
 877         if (ret)
 878                 mlog_errno(ret);
 879         else
 880                 ocfs2_dentry_unlock(dentry, 1);
 881
 882         return ret;
 883 }
 884
 885 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 886 {
 887         if (S_ISDIR(inode->i_mode)) {
 888                 if (inode->i_nlink == 2)
 889                         return 1;
 890                 return 0;
 891         }
 892
 893         if (inode->i_nlink == 1)
 894                 return 1;
 895         return 0;
 896 }
 897
 898 static int ocfs2_unlink(struct inode *dir,
 899                         struct dentry *dentry)
 900 {
 901         int status;
 902         int child_locked = 0;
 903         bool is_unlinkable = false;
 904         struct inode *inode = d_inode(dentry);
 905         struct inode *orphan_dir = NULL;
 906         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 907         u64 blkno;
 908         struct ocfs2_dinode *fe = NULL;
 909         struct buffer_head *fe_bh = NULL;
 910         struct buffer_head *parent_node_bh = NULL;
 911         handle_t *handle = NULL;
 912         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 913         struct ocfs2_dir_lookup_result lookup = { NULL, };
 914         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 915
 916         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 917                            dentry->d_name.name,
 918                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 919                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 920
 921         status = dquot_initialize(dir);
 922         if (status) {
 923                 mlog_errno(status);
 924                 return status;
 925         }
 926
 927         BUG_ON(d_inode(dentry->d_parent) != dir);
 928
 929         if (inode == osb->root_inode)
 930                 return -EPERM;
 931
 932         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 933                                          OI_LS_PARENT);
 934         if (status < 0) {
 935                 if (status != -ENOENT)
 936                         mlog_errno(status);
 937                 return status;
 938         }
 939
 940         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 941                                           dentry->d_name.len, &blkno, dir,
 942                                           &lookup);
 943         if (status < 0) {
 944                 if (status != -ENOENT)
 945                         mlog_errno(status);
 946                 goto leave;
 947         }
 948
 949         if (OCFS2_I(inode)->ip_blkno != blkno) {
 950                 status = -ENOENT;
 951
 952                 trace_ocfs2_unlink_noent(
 953                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 954                                 (unsigned long long)blkno,
 955                                 OCFS2_I(inode)->ip_flags);
 956                 goto leave;
 957         }
 958
 959         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 960         if (status < 0) {
 961                 if (status != -ENOENT)
 962                         mlog_errno(status);
 963                 goto leave;
 964         }
 965         child_locked = 1;
 966
 967         if (S_ISDIR(inode->i_mode)) {
 968                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 969                         status = -ENOTEMPTY;
 970                         goto leave;
 971                 }
 972         }
 973
 974         status = ocfs2_remote_dentry_delete(dentry);
 975         if (status < 0) {
 976                 /* This remote delete should succeed under all normal
 977                  * circumstances. */
 978                 mlog_errno(status);
 979                 goto leave;
 980         }
 981
 982         if (ocfs2_inode_is_unlinkable(inode)) {
 983                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 984                                                   OCFS2_I(inode)->ip_blkno,
 985                                                   orphan_name, &orphan_insert,
 986                                                   false);
 987                 if (status < 0) {
 988                         mlog_errno(status);
 989                         goto leave;
 990                 }
 991                 is_unlinkable = true;
 992         }
 993
 994         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 995         if (IS_ERR(handle)) {
 996                 status = PTR_ERR(handle);
 997                 handle = NULL;
 998                 mlog_errno(status);
 999                 goto leave;
1000         }
1001
1002         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
1003                                          OCFS2_JOURNAL_ACCESS_WRITE);
1004         if (status < 0) {
1005                 mlog_errno(status);
1006                 goto leave;
1007         }
1008
1009         fe = (struct ocfs2_dinode *) fe_bh->b_data;
1010
1011         /* delete the name from the parent dir */
1012         status = ocfs2_delete_entry(handle, dir, &lookup);
1013         if (status < 0) {
1014                 mlog_errno(status);
1015                 goto leave;
1016         }
1017
1018         if (S_ISDIR(inode->i_mode))
1019                 drop_nlink(inode);
1020         drop_nlink(inode);
1021         ocfs2_set_links_count(fe, inode->i_nlink);
1022         ocfs2_journal_dirty(handle, fe_bh);
1023
1024         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1025         if (S_ISDIR(inode->i_mode))
1026                 drop_nlink(dir);
1027
1028         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1029         if (status < 0) {
1030                 mlog_errno(status);
1031                 if (S_ISDIR(inode->i_mode))
1032                         inc_nlink(dir);
1033                 goto leave;
1034         }
1035
1036         if (is_unlinkable) {
1037                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1038                                 orphan_name, &orphan_insert, orphan_dir, false);
1039                 if (status < 0)
1040                         mlog_errno(status);
1041         }
1042
1043 leave:
1044         if (handle)
1045                 ocfs2_commit_trans(osb, handle);
1046
1047         if (orphan_dir) {
1048                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1049                 ocfs2_inode_unlock(orphan_dir, 1);
1050                 mutex_unlock(&orphan_dir->i_mutex);
1051                 iput(orphan_dir);
1052         }
1053
1054         if (child_locked)
1055                 ocfs2_inode_unlock(inode, 1);
1056
1057         ocfs2_inode_unlock(dir, 1);
1058
1059         brelse(fe_bh);
1060         brelse(parent_node_bh);
1061
1062         ocfs2_free_dir_lookup_result(&orphan_insert);
1063         ocfs2_free_dir_lookup_result(&lookup);
1064
1065         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1066                 mlog_errno(status);
1067
1068         return status;
1069 }
1070
1071 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1072                 u64 src_inode_no, u64 dest_inode_no)
1073 {
1074         int ret = 0, i = 0;
1075         u64 parent_inode_no = 0;
1076         u64 child_inode_no = src_inode_no;
1077         struct inode *child_inode;
1078
1079 #define MAX_LOOKUP_TIMES 32
1080         while (1) {
1081                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1082                 if (IS_ERR(child_inode)) {
1083                         ret = PTR_ERR(child_inode);
1084                         break;
1085                 }
1086
1087                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1088                 if (ret < 0) {
1089                         iput(child_inode);
1090                         if (ret != -ENOENT)
1091                                 mlog_errno(ret);
1092                         break;
1093                 }
1094
1095                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1096                                 &parent_inode_no);
1097                 ocfs2_inode_unlock(child_inode, 0);
1098                 iput(child_inode);
1099                 if (ret < 0) {
1100                         ret = -ENOENT;
1101                         break;
1102                 }
1103
1104                 if (parent_inode_no == dest_inode_no) {
1105                         ret = 1;
1106                         break;
1107                 }
1108
1109                 if (parent_inode_no == osb->root_inode->i_ino) {
1110                         ret = 0;
1111                         break;
1112                 }
1113
1114                 child_inode_no = parent_inode_no;
1115
1116                 if (++i >= MAX_LOOKUP_TIMES) {
1117                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1118                                         "may have nested directories, "
1119                                         "src inode: %llu, dest inode: %llu.\n",
1120                                         (unsigned long long)src_inode_no,
1121                                         (unsigned long long)dest_inode_no);
1122                         ret = 0;
1123                         break;
1124                 }
1125         }
1126
1127         return ret;
1128 }
1129
1130 /*
1131  * The only place this should be used is rename and link!
1132  * if they have the same id, then the 1st one is the only one locked.
1133  */
1134 static int ocfs2_double_lock(struct ocfs2_super *osb,
1135                              struct buffer_head **bh1,
1136                              struct inode *inode1,
1137                              struct buffer_head **bh2,
1138                              struct inode *inode2,
1139                              int rename)
1140 {
1141         int status;
1142         int inode1_is_ancestor, inode2_is_ancestor;
1143         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1144         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1145
1146         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1147                                 (unsigned long long)oi2->ip_blkno);
1148
1149         if (*bh1)
1150                 *bh1 = NULL;
1151         if (*bh2)
1152                 *bh2 = NULL;
1153
1154         /* we always want to lock the one with the lower lockid first.
1155          * and if they are nested, we lock ancestor first */
1156         if (oi1->ip_blkno != oi2->ip_blkno) {
1157                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1158                                 oi1->ip_blkno);
1159                 if (inode1_is_ancestor < 0) {
1160                         status = inode1_is_ancestor;
1161                         goto bail;
1162                 }
1163
1164                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1165                                 oi2->ip_blkno);
1166                 if (inode2_is_ancestor < 0) {
1167                         status = inode2_is_ancestor;
1168                         goto bail;
1169                 }
1170
1171                 if ((inode1_is_ancestor == 1) ||
1172                                 (oi1->ip_blkno < oi2->ip_blkno &&
1173                                 inode2_is_ancestor == 0)) {
1174                         /* switch id1 and id2 around */
1175                         swap(bh2, bh1);
1176                         swap(inode2, inode1);
1177                 }
1178                 /* lock id2 */
1179                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1180                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1181                 if (status < 0) {
1182                         if (status != -ENOENT)
1183                                 mlog_errno(status);
1184                         goto bail;
1185                 }
1186         }
1187
1188         /* lock id1 */
1189         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1190                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1191         if (status < 0) {
1192                 /*
1193                  * An error return must mean that no cluster locks
1194                  * were held on function exit.
1195                  */
1196                 if (oi1->ip_blkno != oi2->ip_blkno) {
1197                         ocfs2_inode_unlock(inode2, 1);
1198                         brelse(*bh2);
1199                         *bh2 = NULL;
1200                 }
1201
1202                 if (status != -ENOENT)
1203                         mlog_errno(status);
1204         }
1205
1206         trace_ocfs2_double_lock_end(
1207                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1208                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1209
1210 bail:
1211         if (status)
1212                 mlog_errno(status);
1213         return status;
1214 }
1215
1216 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1217 {
1218         ocfs2_inode_unlock(inode1, 1);
1219
1220         if (inode1 != inode2)
1221                 ocfs2_inode_unlock(inode2, 1);
1222 }
1223
1224 static int ocfs2_rename(struct inode *old_dir,
1225                         struct dentry *old_dentry,
1226                         struct inode *new_dir,
1227                         struct dentry *new_dentry)
1228 {
1229         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1230         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1231         struct inode *old_inode = d_inode(old_dentry);
1232         struct inode *new_inode = d_inode(new_dentry);
1233         struct inode *orphan_dir = NULL;
1234         struct ocfs2_dinode *newfe = NULL;
1235         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1236         struct buffer_head *newfe_bh = NULL;
1237         struct buffer_head *old_inode_bh = NULL;
1238         struct ocfs2_super *osb = NULL;
1239         u64 newfe_blkno, old_de_ino;
1240         handle_t *handle = NULL;
1241         struct buffer_head *old_dir_bh = NULL;
1242         struct buffer_head *new_dir_bh = NULL;
1243         u32 old_dir_nlink = old_dir->i_nlink;
1244         struct ocfs2_dinode *old_di;
1245         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1246         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1247         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1248         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1249         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1250         bool should_add_orphan = false;
1251
1252         /* At some point it might be nice to break this function up a
1253          * bit. */
1254
1255         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1256                            old_dentry->d_name.len, old_dentry->d_name.name,
1257                            new_dentry->d_name.len, new_dentry->d_name.name);
1258
1259         status = dquot_initialize(old_dir);
1260         if (status) {
1261                 mlog_errno(status);
1262                 goto bail;
1263         }
1264         status = dquot_initialize(new_dir);
1265         if (status) {
1266                 mlog_errno(status);
1267                 goto bail;
1268         }
1269
1270         osb = OCFS2_SB(old_dir->i_sb);
1271
1272         if (new_inode) {
1273                 if (!igrab(new_inode))
1274                         BUG();
1275         }
1276
1277         /* Assume a directory hierarchy thusly:
1278          * a/b/c
1279          * a/d
1280          * a,b,c, and d are all directories.
1281          *
1282          * from cwd of 'a' on both nodes:
1283          * node1: mv b/c d
1284          * node2: mv d   b/c
1285          *
1286          * And that's why, just like the VFS, we need a file system
1287          * rename lock. */
1288         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1289                 status = ocfs2_rename_lock(osb);
1290                 if (status < 0) {
1291                         mlog_errno(status);
1292                         goto bail;
1293                 }
1294                 rename_lock = 1;
1295
1296                 /* here we cannot guarantee the inodes haven't just been
1297                  * changed, so check if they are nested again */
1298                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1299                                 old_inode->i_ino);
1300                 if (status < 0) {
1301                         mlog_errno(status);
1302                         goto bail;
1303                 } else if (status == 1) {
1304                         status = -EPERM;
1305                         trace_ocfs2_rename_not_permitted(
1306                                         (unsigned long long)old_inode->i_ino,
1307                                         (unsigned long long)new_dir->i_ino);
1308                         goto bail;
1309                 }
1310         }
1311
1312         /* if old and new are the same, this'll just do one lock. */
1313         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1314                                    &new_dir_bh, new_dir, 1);
1315         if (status < 0) {
1316                 mlog_errno(status);
1317                 goto bail;
1318         }
1319         parents_locked = 1;
1320
1321         if (!new_dir->i_nlink) {
1322                 status = -EACCES;
1323                 goto bail;
1324         }
1325
1326         /* make sure both dirs have bhs
1327          * get an extra ref on old_dir_bh if old==new */
1328         if (!new_dir_bh) {
1329                 if (old_dir_bh) {
1330                         new_dir_bh = old_dir_bh;
1331                         get_bh(new_dir_bh);
1332                 } else {
1333                         mlog(ML_ERROR, "no old_dir_bh!\n");
1334                         status = -EIO;
1335                         goto bail;
1336                 }
1337         }
1338
1339         /*
1340          * Aside from allowing a meta data update, the locking here
1341          * also ensures that the downconvert thread on other nodes
1342          * won't have to concurrently downconvert the inode and the
1343          * dentry locks.
1344          */
1345         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1346                                          OI_LS_PARENT);
1347         if (status < 0) {
1348                 if (status != -ENOENT)
1349                         mlog_errno(status);
1350                 goto bail;
1351         }
1352         old_child_locked = 1;
1353
1354         status = ocfs2_remote_dentry_delete(old_dentry);
1355         if (status < 0) {
1356                 mlog_errno(status);
1357                 goto bail;
1358         }
1359
1360         if (S_ISDIR(old_inode->i_mode)) {
1361                 u64 old_inode_parent;
1362
1363                 update_dot_dot = 1;
1364                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1365                                                   old_inode,
1366                                                   &old_inode_dot_dot_res);
1367                 if (status) {
1368                         status = -EIO;
1369                         goto bail;
1370                 }
1371
1372                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1373                         status = -EIO;
1374                         goto bail;
1375                 }
1376
1377                 if (!new_inode && new_dir != old_dir &&
1378                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1379                         status = -EMLINK;
1380                         goto bail;
1381                 }
1382         }
1383
1384         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1385                                             old_dentry->d_name.len,
1386                                             &old_de_ino);
1387         if (status) {
1388                 status = -ENOENT;
1389                 goto bail;
1390         }
1391
1392         /*
1393          *  Check for inode number is _not_ due to possible IO errors.
1394          *  We might rmdir the source, keep it as pwd of some process
1395          *  and merrily kill the link to whatever was created under the
1396          *  same name. Goodbye sticky bit ;-<
1397          */
1398         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1399                 status = -ENOENT;
1400                 goto bail;
1401         }
1402
1403         /* check if the target already exists (in which case we need
1404          * to delete it */
1405         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1406                                           new_dentry->d_name.len,
1407                                           &newfe_blkno, new_dir,
1408                                           &target_lookup_res);
1409         /* The only error we allow here is -ENOENT because the new
1410          * file not existing is perfectly valid. */
1411         if ((status < 0) && (status != -ENOENT)) {
1412                 /* If we cannot find the file specified we should just */
1413                 /* return the error... */
1414                 mlog_errno(status);
1415                 goto bail;
1416         }
1417         if (status == 0)
1418                 target_exists = 1;
1419
1420         if (!target_exists && new_inode) {
1421                 /*
1422                  * Target was unlinked by another node while we were
1423                  * waiting to get to ocfs2_rename(). There isn't
1424                  * anything we can do here to help the situation, so
1425                  * bubble up the appropriate error.
1426                  */
1427                 status = -ENOENT;
1428                 goto bail;
1429         }
1430
1431         /* In case we need to overwrite an existing file, we blow it
1432          * away first */
1433         if (target_exists) {
1434                 /* VFS didn't think there existed an inode here, but
1435                  * someone else in the cluster must have raced our
1436                  * rename to create one. Today we error cleanly, in
1437                  * the future we should consider calling iget to build
1438                  * a new struct inode for this entry. */
1439                 if (!new_inode) {
1440                         status = -EACCES;
1441
1442                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1443                                                 new_dentry->d_name.name);
1444                         goto bail;
1445                 }
1446
1447                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1448                         status = -EACCES;
1449
1450                         trace_ocfs2_rename_disagree(
1451                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1452                              (unsigned long long)newfe_blkno,
1453                              OCFS2_I(new_inode)->ip_flags);
1454                         goto bail;
1455                 }
1456
1457                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1458                 if (status < 0) {
1459                         if (status != -ENOENT)
1460                                 mlog_errno(status);
1461                         goto bail;
1462                 }
1463                 new_child_locked = 1;
1464
1465                 status = ocfs2_remote_dentry_delete(new_dentry);
1466                 if (status < 0) {
1467                         mlog_errno(status);
1468                         goto bail;
1469                 }
1470
1471                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1472
1473                 trace_ocfs2_rename_over_existing(
1474                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1475                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1476
1477                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1478                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1479                                                 OCFS2_I(new_inode)->ip_blkno,
1480                                                 orphan_name, &orphan_insert,
1481                                                 false);
1482                         if (status < 0) {
1483                                 mlog_errno(status);
1484                                 goto bail;
1485                         }
1486                         should_add_orphan = true;
1487                 }
1488         } else {
1489                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1490
1491                 status = ocfs2_check_dir_for_entry(new_dir,
1492                                                    new_dentry->d_name.name,
1493                                                    new_dentry->d_name.len);
1494                 if (status)
1495                         goto bail;
1496
1497                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1498                                                       new_dentry->d_name.name,
1499                                                       new_dentry->d_name.len,
1500                                                       &target_insert);
1501                 if (status < 0) {
1502                         mlog_errno(status);
1503                         goto bail;
1504                 }
1505         }
1506
1507         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1508         if (IS_ERR(handle)) {
1509                 status = PTR_ERR(handle);
1510                 handle = NULL;
1511                 mlog_errno(status);
1512                 goto bail;
1513         }
1514
1515         if (target_exists) {
1516                 if (S_ISDIR(new_inode->i_mode)) {
1517                         if (new_inode->i_nlink != 2 ||
1518                             !ocfs2_empty_dir(new_inode)) {
1519                                 status = -ENOTEMPTY;
1520                                 goto bail;
1521                         }
1522                 }
1523                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1524                                                  newfe_bh,
1525                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1526                 if (status < 0) {
1527                         mlog_errno(status);
1528                         goto bail;
1529                 }
1530
1531                 /* change the dirent to point to the correct inode */
1532                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1533                                             old_inode);
1534                 if (status < 0) {
1535                         mlog_errno(status);
1536                         goto bail;
1537                 }
1538                 new_dir->i_version++;
1539
1540                 if (S_ISDIR(new_inode->i_mode))
1541                         ocfs2_set_links_count(newfe, 0);
1542                 else
1543                         ocfs2_add_links_count(newfe, -1);
1544                 ocfs2_journal_dirty(handle, newfe_bh);
1545                 if (should_add_orphan) {
1546                         status = ocfs2_orphan_add(osb, handle, new_inode,
1547                                         newfe_bh, orphan_name,
1548                                         &orphan_insert, orphan_dir, false);
1549                         if (status < 0) {
1550                                 mlog_errno(status);
1551                                 goto bail;
1552                         }
1553                 }
1554         } else {
1555                 /* if the name was not found in new_dir, add it now */
1556                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1557                                          OCFS2_I(old_inode)->ip_blkno,
1558                                          new_dir_bh, &target_insert);
1559         }
1560
1561         old_inode->i_ctime = CURRENT_TIME;
1562         mark_inode_dirty(old_inode);
1563
1564         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1565                                          old_inode_bh,
1566                                          OCFS2_JOURNAL_ACCESS_WRITE);
1567         if (status >= 0) {
1568                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1569
1570                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1571                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1572                 ocfs2_journal_dirty(handle, old_inode_bh);
1573         } else
1574                 mlog_errno(status);
1575
1576         /*
1577          * Now that the name has been added to new_dir, remove the old name.
1578          *
1579          * We don't keep any directory entry context around until now
1580          * because the insert might have changed the type of directory
1581          * we're dealing with.
1582          */
1583         status = ocfs2_find_entry(old_dentry->d_name.name,
1584                                   old_dentry->d_name.len, old_dir,
1585                                   &old_entry_lookup);
1586         if (status) {
1587                 if (!is_journal_aborted(osb->journal->j_journal)) {
1588                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1589                                         "is not deleted.",
1590                                         new_dentry->d_name.len, new_dentry->d_name.name,
1591                                         old_dentry->d_name.len, old_dentry->d_name.name);
1592                 }
1593                 goto bail;
1594         }
1595
1596         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1597         if (status < 0) {
1598                 mlog_errno(status);
1599                 if (!is_journal_aborted(osb->journal->j_journal)) {
1600                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1601                                         "is not deleted.",
1602                                         new_dentry->d_name.len, new_dentry->d_name.name,
1603                                         old_dentry->d_name.len, old_dentry->d_name.name);
1604                 }
1605                 goto bail;
1606         }
1607
1608         if (new_inode) {
1609                 drop_nlink(new_inode);
1610                 new_inode->i_ctime = CURRENT_TIME;
1611         }
1612         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1613
1614         if (update_dot_dot) {
1615                 status = ocfs2_update_entry(old_inode, handle,
1616                                             &old_inode_dot_dot_res, new_dir);
1617                 drop_nlink(old_dir);
1618                 if (new_inode) {
1619                         drop_nlink(new_inode);
1620                 } else {
1621                         inc_nlink(new_dir);
1622                         mark_inode_dirty(new_dir);
1623                 }
1624         }
1625         mark_inode_dirty(old_dir);
1626         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1627         if (new_inode) {
1628                 mark_inode_dirty(new_inode);
1629                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1630         }
1631
1632         if (old_dir != new_dir) {
1633                 /* Keep the same times on both directories.*/
1634                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1635
1636                 /*
1637                  * This will also pick up the i_nlink change from the
1638                  * block above.
1639                  */
1640                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1641         }
1642
1643         if (old_dir_nlink != old_dir->i_nlink) {
1644                 if (!old_dir_bh) {
1645                         mlog(ML_ERROR, "need to change nlink for old dir "
1646                              "%llu from %d to %d but bh is NULL!\n",
1647                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1648                              (int)old_dir_nlink, old_dir->i_nlink);
1649                 } else {
1650                         struct ocfs2_dinode *fe;
1651                         status = ocfs2_journal_access_di(handle,
1652                                                          INODE_CACHE(old_dir),
1653                                                          old_dir_bh,
1654                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1655                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1656                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1657                         ocfs2_journal_dirty(handle, old_dir_bh);
1658                 }
1659         }
1660         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1661         status = 0;
1662 bail:
1663         if (handle)
1664                 ocfs2_commit_trans(osb, handle);
1665
1666         if (orphan_dir) {
1667                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1668                 ocfs2_inode_unlock(orphan_dir, 1);
1669                 mutex_unlock(&orphan_dir->i_mutex);
1670                 iput(orphan_dir);
1671         }
1672
1673         if (new_child_locked)
1674                 ocfs2_inode_unlock(new_inode, 1);
1675
1676         if (old_child_locked)
1677                 ocfs2_inode_unlock(old_inode, 1);
1678
1679         if (parents_locked)
1680                 ocfs2_double_unlock(old_dir, new_dir);
1681
1682         if (rename_lock)
1683                 ocfs2_rename_unlock(osb);
1684
1685         if (new_inode)
1686                 sync_mapping_buffers(old_inode->i_mapping);
1687
1688         if (new_inode)
1689                 iput(new_inode);
1690
1691         ocfs2_free_dir_lookup_result(&target_lookup_res);
1692         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1693         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1694         ocfs2_free_dir_lookup_result(&orphan_insert);
1695         ocfs2_free_dir_lookup_result(&target_insert);
1696
1697         brelse(newfe_bh);
1698         brelse(old_inode_bh);
1699         brelse(old_dir_bh);
1700         brelse(new_dir_bh);
1701
1702         if (status)
1703                 mlog_errno(status);
1704
1705         return status;
1706 }
1707
1708 /*
1709  * we expect i_size = strlen(symname). Copy symname into the file
1710  * data, including the null terminator.
1711  */
1712 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1713                                      handle_t *handle,
1714                                      struct inode *inode,
1715                                      const char *symname)
1716 {
1717         struct buffer_head **bhs = NULL;
1718         const char *c;
1719         struct super_block *sb = osb->sb;
1720         u64 p_blkno, p_blocks;
1721         int virtual, blocks, status, i, bytes_left;
1722
1723         bytes_left = i_size_read(inode) + 1;
1724         /* we can't trust i_blocks because we're actually going to
1725          * write i_size + 1 bytes. */
1726         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1727
1728         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1729                                         i_size_read(inode), blocks);
1730
1731         /* Sanity check -- make sure we're going to fit. */
1732         if (bytes_left >
1733             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1734                 status = -EIO;
1735                 mlog_errno(status);
1736                 goto bail;
1737         }
1738
1739         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1740         if (!bhs) {
1741                 status = -ENOMEM;
1742                 mlog_errno(status);
1743                 goto bail;
1744         }
1745
1746         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1747                                              NULL);
1748         if (status < 0) {
1749                 mlog_errno(status);
1750                 goto bail;
1751         }
1752
1753         /* links can never be larger than one cluster so we know this
1754          * is all going to be contiguous, but do a sanity check
1755          * anyway. */
1756         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1757                 status = -EIO;
1758                 mlog_errno(status);
1759                 goto bail;
1760         }
1761
1762         virtual = 0;
1763         while(bytes_left > 0) {
1764                 c = &symname[virtual * sb->s_blocksize];
1765
1766                 bhs[virtual] = sb_getblk(sb, p_blkno);
1767                 if (!bhs[virtual]) {
1768                         status = -ENOMEM;
1769                         mlog_errno(status);
1770                         goto bail;
1771                 }
1772                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1773                                               bhs[virtual]);
1774
1775                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1776                                               bhs[virtual],
1777                                               OCFS2_JOURNAL_ACCESS_CREATE);
1778                 if (status < 0) {
1779                         mlog_errno(status);
1780                         goto bail;
1781                 }
1782
1783                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1784
1785                 memcpy(bhs[virtual]->b_data, c,
1786                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1787                        bytes_left);
1788
1789                 ocfs2_journal_dirty(handle, bhs[virtual]);
1790
1791                 virtual++;
1792                 p_blkno++;
1793                 bytes_left -= sb->s_blocksize;
1794         }
1795
1796         status = 0;
1797 bail:
1798
1799         if (bhs) {
1800                 for(i = 0; i < blocks; i++)
1801                         brelse(bhs[i]);
1802                 kfree(bhs);
1803         }
1804
1805         if (status)
1806                 mlog_errno(status);
1807         return status;
1808 }
1809
1810 static int ocfs2_symlink(struct inode *dir,
1811                          struct dentry *dentry,
1812                          const char *symname)
1813 {
1814         int status, l, credits;
1815         u64 newsize;
1816         struct ocfs2_super *osb = NULL;
1817         struct inode *inode = NULL;
1818         struct super_block *sb;
1819         struct buffer_head *new_fe_bh = NULL;
1820         struct buffer_head *parent_fe_bh = NULL;
1821         struct ocfs2_dinode *fe = NULL;
1822         struct ocfs2_dinode *dirfe;
1823         handle_t *handle = NULL;
1824         struct ocfs2_alloc_context *inode_ac = NULL;
1825         struct ocfs2_alloc_context *data_ac = NULL;
1826         struct ocfs2_alloc_context *xattr_ac = NULL;
1827         int want_clusters = 0;
1828         int xattr_credits = 0;
1829         struct ocfs2_security_xattr_info si = {
1830                 .enable = 1,
1831         };
1832         int did_quota = 0, did_quota_inode = 0;
1833         struct ocfs2_dir_lookup_result lookup = { NULL, };
1834         sigset_t oldset;
1835         int did_block_signals = 0;
1836         struct ocfs2_dentry_lock *dl = NULL;
1837
1838         trace_ocfs2_symlink_begin(dir, dentry, symname,
1839                                   dentry->d_name.len, dentry->d_name.name);
1840
1841         status = dquot_initialize(dir);
1842         if (status) {
1843                 mlog_errno(status);
1844                 goto bail;
1845         }
1846
1847         sb = dir->i_sb;
1848         osb = OCFS2_SB(sb);
1849
1850         l = strlen(symname) + 1;
1851
1852         credits = ocfs2_calc_symlink_credits(sb);
1853
1854         /* lock the parent directory */
1855         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1856         if (status < 0) {
1857                 if (status != -ENOENT)
1858                         mlog_errno(status);
1859                 return status;
1860         }
1861
1862         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1863         if (!ocfs2_read_links_count(dirfe)) {
1864                 /* can't make a file in a deleted directory. */
1865                 status = -ENOENT;
1866                 goto bail;
1867         }
1868
1869         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1870                                            dentry->d_name.len);
1871         if (status)
1872                 goto bail;
1873
1874         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1875                                               dentry->d_name.name,
1876                                               dentry->d_name.len, &lookup);
1877         if (status < 0) {
1878                 mlog_errno(status);
1879                 goto bail;
1880         }
1881
1882         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1883         if (status < 0) {
1884                 if (status != -ENOSPC)
1885                         mlog_errno(status);
1886                 goto bail;
1887         }
1888
1889         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1890         if (IS_ERR(inode)) {
1891                 status = PTR_ERR(inode);
1892                 inode = NULL;
1893                 mlog_errno(status);
1894                 goto bail;
1895         }
1896
1897         /* get security xattr */
1898         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1899         if (status) {
1900                 if (status == -EOPNOTSUPP)
1901                         si.enable = 0;
1902                 else {
1903                         mlog_errno(status);
1904                         goto bail;
1905                 }
1906         }
1907
1908         /* calculate meta data/clusters for setting security xattr */
1909         if (si.enable) {
1910                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1911                                                   &xattr_credits, &xattr_ac);
1912                 if (status < 0) {
1913                         mlog_errno(status);
1914                         goto bail;
1915                 }
1916         }
1917
1918         /* don't reserve bitmap space for fast symlinks. */
1919         if (l > ocfs2_fast_symlink_chars(sb))
1920                 want_clusters += 1;
1921
1922         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1923         if (status < 0) {
1924                 if (status != -ENOSPC)
1925                         mlog_errno(status);
1926                 goto bail;
1927         }
1928
1929         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1930         if (IS_ERR(handle)) {
1931                 status = PTR_ERR(handle);
1932                 handle = NULL;
1933                 mlog_errno(status);
1934                 goto bail;
1935         }
1936
1937         /* Starting to change things, restart is no longer possible. */
1938         ocfs2_block_signals(&oldset);
1939         did_block_signals = 1;
1940
1941         status = dquot_alloc_inode(inode);
1942         if (status)
1943                 goto bail;
1944         did_quota_inode = 1;
1945
1946         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1947                                    dentry->d_name.name,
1948                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1949                                    inode->i_mode);
1950
1951         status = ocfs2_mknod_locked(osb, dir, inode,
1952                                     0, &new_fe_bh, parent_fe_bh, handle,
1953                                     inode_ac);
1954         if (status < 0) {
1955                 mlog_errno(status);
1956                 goto bail;
1957         }
1958
1959         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1960         inode->i_rdev = 0;
1961         newsize = l - 1;
1962         inode->i_op = &ocfs2_symlink_inode_operations;
1963         inode_nohighmem(inode);
1964         if (l > ocfs2_fast_symlink_chars(sb)) {
1965                 u32 offset = 0;
1966
1967                 status = dquot_alloc_space_nodirty(inode,
1968                     ocfs2_clusters_to_bytes(osb->sb, 1));
1969                 if (status)
1970                         goto bail;
1971                 did_quota = 1;
1972                 inode->i_mapping->a_ops = &ocfs2_aops;
1973                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1974                                               new_fe_bh,
1975                                               handle, data_ac, NULL,
1976                                               NULL);
1977                 if (status < 0) {
1978                         if (status != -ENOSPC && status != -EINTR) {
1979                                 mlog(ML_ERROR,
1980                                      "Failed to extend file to %llu\n",
1981                                      (unsigned long long)newsize);
1982                                 mlog_errno(status);
1983                                 status = -ENOSPC;
1984                         }
1985                         goto bail;
1986                 }
1987                 i_size_write(inode, newsize);
1988                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1989         } else {
1990                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1991                 memcpy((char *) fe->id2.i_symlink, symname, l);
1992                 i_size_write(inode, newsize);
1993                 inode->i_blocks = 0;
1994         }
1995
1996         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1997         if (status < 0) {
1998                 mlog_errno(status);
1999                 goto bail;
2000         }
2001
2002         if (!ocfs2_inode_is_fast_symlink(inode)) {
2003                 status = ocfs2_create_symlink_data(osb, handle, inode,
2004                                                    symname);
2005                 if (status < 0) {
2006                         mlog_errno(status);
2007                         goto bail;
2008                 }
2009         }
2010
2011         if (si.enable) {
2012                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
2013                                                  xattr_ac, data_ac);
2014                 if (status < 0) {
2015                         mlog_errno(status);
2016                         goto bail;
2017                 }
2018         }
2019
2020         /*
2021          * Do this before adding the entry to the directory. We add
2022          * also set d_op after success so that ->d_iput() will cleanup
2023          * the dentry lock even if ocfs2_add_entry() fails below.
2024          */
2025         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2026         if (status) {
2027                 mlog_errno(status);
2028                 goto bail;
2029         }
2030
2031         dl = dentry->d_fsdata;
2032
2033         status = ocfs2_add_entry(handle, dentry, inode,
2034                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2035                                  &lookup);
2036         if (status < 0) {
2037                 mlog_errno(status);
2038                 goto bail;
2039         }
2040
2041         insert_inode_hash(inode);
2042         d_instantiate(dentry, inode);
2043 bail:
2044         if (status < 0 && did_quota)
2045                 dquot_free_space_nodirty(inode,
2046                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2047         if (status < 0 && did_quota_inode)
2048                 dquot_free_inode(inode);
2049         if (handle)
2050                 ocfs2_commit_trans(osb, handle);
2051
2052         ocfs2_inode_unlock(dir, 1);
2053         if (did_block_signals)
2054                 ocfs2_unblock_signals(&oldset);
2055
2056         brelse(new_fe_bh);
2057         brelse(parent_fe_bh);
2058         kfree(si.value);
2059         ocfs2_free_dir_lookup_result(&lookup);
2060         if (inode_ac)
2061                 ocfs2_free_alloc_context(inode_ac);
2062         if (data_ac)
2063                 ocfs2_free_alloc_context(data_ac);
2064         if (xattr_ac)
2065                 ocfs2_free_alloc_context(xattr_ac);
2066         if ((status < 0) && inode) {
2067                 if (dl)
2068                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2069
2070                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2071                 clear_nlink(inode);
2072                 iput(inode);
2073         }
2074
2075         if (status)
2076                 mlog_errno(status);
2077
2078         return status;
2079 }
2080
2081 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2082 {
2083         int status, namelen;
2084
2085         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2086                            (long long)blkno);
2087         if (namelen <= 0) {
2088                 if (namelen)
2089                         status = namelen;
2090                 else
2091                         status = -EINVAL;
2092                 mlog_errno(status);
2093                 goto bail;
2094         }
2095         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2096                 status = -EINVAL;
2097                 mlog_errno(status);
2098                 goto bail;
2099         }
2100
2101         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2102
2103         status = 0;
2104 bail:
2105         if (status < 0)
2106                 mlog_errno(status);
2107         return status;
2108 }
2109
2110 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2111                                         struct inode **ret_orphan_dir,
2112                                         struct buffer_head **ret_orphan_dir_bh)
2113 {
2114         struct inode *orphan_dir_inode;
2115         struct buffer_head *orphan_dir_bh = NULL;
2116         int ret = 0;
2117
2118         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2119                                                        ORPHAN_DIR_SYSTEM_INODE,
2120                                                        osb->slot_num);
2121         if (!orphan_dir_inode) {
2122                 ret = -ENOENT;
2123                 mlog_errno(ret);
2124                 return ret;
2125         }
2126
2127         mutex_lock(&orphan_dir_inode->i_mutex);
2128
2129         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2130         if (ret < 0) {
2131                 mutex_unlock(&orphan_dir_inode->i_mutex);
2132                 iput(orphan_dir_inode);
2133
2134                 mlog_errno(ret);
2135                 return ret;
2136         }
2137
2138         *ret_orphan_dir = orphan_dir_inode;
2139         *ret_orphan_dir_bh = orphan_dir_bh;
2140
2141         return 0;
2142 }
2143
2144 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2145                                       struct buffer_head *orphan_dir_bh,
2146                                       u64 blkno,
2147                                       char *name,
2148                                       struct ocfs2_dir_lookup_result *lookup,
2149                                       bool dio)
2150 {
2151         int ret;
2152         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2153         int namelen = dio ?
2154                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2155                         OCFS2_ORPHAN_NAMELEN;
2156
2157         if (dio) {
2158                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2159                                 OCFS2_DIO_ORPHAN_PREFIX);
2160                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2161                         ret = -EINVAL;
2162                         mlog_errno(ret);
2163                         return ret;
2164                 }
2165
2166                 ret = ocfs2_blkno_stringify(blkno,
2167                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2168         } else
2169                 ret = ocfs2_blkno_stringify(blkno, name);
2170         if (ret < 0) {
2171                 mlog_errno(ret);
2172                 return ret;
2173         }
2174
2175         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2176                                            orphan_dir_bh, name,
2177                                            namelen, lookup);
2178         if (ret < 0) {
2179                 mlog_errno(ret);
2180                 return ret;
2181         }
2182
2183         return 0;
2184 }
2185
2186 /**
2187  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2188  * insertion of an orphan.
2189  * @osb: ocfs2 file system
2190  * @ret_orphan_dir: Orphan dir inode - returned locked!
2191  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2192  * @lookup: dir lookup result, to be passed back into functions like
2193  *          ocfs2_orphan_add
2194  *
2195  * Returns zero on success and the ret_orphan_dir, name and lookup
2196  * fields will be populated.
2197  *
2198  * Returns non-zero on failure.
2199  */
2200 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2201                                     struct inode **ret_orphan_dir,
2202                                     u64 blkno,
2203                                     char *name,
2204                                     struct ocfs2_dir_lookup_result *lookup,
2205                                     bool dio)
2206 {
2207         struct inode *orphan_dir_inode = NULL;
2208         struct buffer_head *orphan_dir_bh = NULL;
2209         int ret = 0;
2210
2211         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2212                                            &orphan_dir_bh);
2213         if (ret < 0) {
2214                 mlog_errno(ret);
2215                 return ret;
2216         }
2217
2218         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2219                                          blkno, name, lookup, dio);
2220         if (ret < 0) {
2221                 mlog_errno(ret);
2222                 goto out;
2223         }
2224
2225         *ret_orphan_dir = orphan_dir_inode;
2226
2227 out:
2228         brelse(orphan_dir_bh);
2229
2230         if (ret) {
2231                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2232                 mutex_unlock(&orphan_dir_inode->i_mutex);
2233                 iput(orphan_dir_inode);
2234         }
2235
2236         if (ret)
2237                 mlog_errno(ret);
2238         return ret;
2239 }
2240
2241 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2242                             handle_t *handle,
2243                             struct inode *inode,
2244                             struct buffer_head *fe_bh,
2245                             char *name,
2246                             struct ocfs2_dir_lookup_result *lookup,
2247                             struct inode *orphan_dir_inode,
2248                             bool dio)
2249 {
2250         struct buffer_head *orphan_dir_bh = NULL;
2251         int status = 0;
2252         struct ocfs2_dinode *orphan_fe;
2253         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2254         int namelen = dio ?
2255                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2256                         OCFS2_ORPHAN_NAMELEN;
2257
2258         trace_ocfs2_orphan_add_begin(
2259                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2260
2261         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2262         if (status < 0) {
2263                 mlog_errno(status);
2264                 goto leave;
2265         }
2266
2267         status = ocfs2_journal_access_di(handle,
2268                                          INODE_CACHE(orphan_dir_inode),
2269                                          orphan_dir_bh,
2270                                          OCFS2_JOURNAL_ACCESS_WRITE);
2271         if (status < 0) {
2272                 mlog_errno(status);
2273                 goto leave;
2274         }
2275
2276         /*
2277          * We're going to journal the change of i_flags and i_orphaned_slot.
2278          * It's safe anyway, though some callers may duplicate the journaling.
2279          * Journaling within the func just make the logic look more
2280          * straightforward.
2281          */
2282         status = ocfs2_journal_access_di(handle,
2283                                          INODE_CACHE(inode),
2284                                          fe_bh,
2285                                          OCFS2_JOURNAL_ACCESS_WRITE);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto leave;
2289         }
2290
2291         /* we're a cluster, and nlink can change on disk from
2292          * underneath us... */
2293         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2294         if (S_ISDIR(inode->i_mode))
2295                 ocfs2_add_links_count(orphan_fe, 1);
2296         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2297         ocfs2_journal_dirty(handle, orphan_dir_bh);
2298
2299         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2300                                    namelen, inode,
2301                                    OCFS2_I(inode)->ip_blkno,
2302                                    orphan_dir_bh, lookup);
2303         if (status < 0) {
2304                 mlog_errno(status);
2305                 goto rollback;
2306         }
2307
2308         if (dio) {
2309                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2310                  * slot.
2311                  */
2312                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2313                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2314         } else {
2315                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2316                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2317
2318                 /* Record which orphan dir our inode now resides
2319                  * in. delete_inode will use this to determine which orphan
2320                  * dir to lock. */
2321                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2322         }
2323
2324         ocfs2_journal_dirty(handle, fe_bh);
2325
2326         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2327                                    osb->slot_num);
2328
2329 rollback:
2330         if (status < 0) {
2331                 if (S_ISDIR(inode->i_mode))
2332                         ocfs2_add_links_count(orphan_fe, -1);
2333                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2334         }
2335
2336 leave:
2337         brelse(orphan_dir_bh);
2338
2339         return status;
2340 }
2341
2342 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2343 int ocfs2_orphan_del(struct ocfs2_super *osb,
2344                      handle_t *handle,
2345                      struct inode *orphan_dir_inode,
2346                      struct inode *inode,
2347                      struct buffer_head *orphan_dir_bh,
2348                      bool dio)
2349 {
2350         const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2351         char name[namelen + 1];
2352         struct ocfs2_dinode *orphan_fe;
2353         int status = 0;
2354         struct ocfs2_dir_lookup_result lookup = { NULL, };
2355
2356         if (dio) {
2357                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2358                                 OCFS2_DIO_ORPHAN_PREFIX);
2359                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2360                         status = -EINVAL;
2361                         mlog_errno(status);
2362                         return status;
2363                 }
2364
2365                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2366                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2367         } else
2368                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2369         if (status < 0) {
2370                 mlog_errno(status);
2371                 goto leave;
2372         }
2373
2374         trace_ocfs2_orphan_del(
2375              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2376              name, strlen(name));
2377
2378         /* find it's spot in the orphan directory */
2379         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2380                                   &lookup);
2381         if (status) {
2382                 mlog_errno(status);
2383                 goto leave;
2384         }
2385
2386         /* remove it from the orphan directory */
2387         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2388         if (status < 0) {
2389                 mlog_errno(status);
2390                 goto leave;
2391         }
2392
2393         status = ocfs2_journal_access_di(handle,
2394                                          INODE_CACHE(orphan_dir_inode),
2395                                          orphan_dir_bh,
2396                                          OCFS2_JOURNAL_ACCESS_WRITE);
2397         if (status < 0) {
2398                 mlog_errno(status);
2399                 goto leave;
2400         }
2401
2402         /* do the i_nlink dance! :) */
2403         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2404         if (S_ISDIR(inode->i_mode))
2405                 ocfs2_add_links_count(orphan_fe, -1);
2406         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2407         ocfs2_journal_dirty(handle, orphan_dir_bh);
2408
2409 leave:
2410         ocfs2_free_dir_lookup_result(&lookup);
2411
2412         if (status)
2413                 mlog_errno(status);
2414         return status;
2415 }
2416
2417 /**
2418  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2419  * allocated file. This is different from the typical 'add to orphan dir'
2420  * operation in that the inode does not yet exist. This is a problem because
2421  * the orphan dir stringifies the inode block number to come up with it's
2422  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2423  * problem. This function works around it by calling deeper into the orphan
2424  * and suballoc code than other callers. Use this only by necessity.
2425  * @dir: The directory which this inode will ultimately wind up under - not the
2426  * orphan dir!
2427  * @dir_bh: buffer_head the @dir inode block
2428  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2429  * with the string to be used for orphan dirent. Pass back to the orphan dir
2430  * code.
2431  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2432  * dir code.
2433  * @ret_di_blkno: block number where the new inode will be allocated.
2434  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2435  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2436  *
2437  * Returns zero on success and the ret_orphan_dir, name and lookup
2438  * fields will be populated.
2439  *
2440  * Returns non-zero on failure.
2441  */
2442 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2443                                         struct buffer_head *dir_bh,
2444                                         char *orphan_name,
2445                                         struct inode **ret_orphan_dir,
2446                                         u64 *ret_di_blkno,
2447                                         struct ocfs2_dir_lookup_result *orphan_insert,
2448                                         struct ocfs2_alloc_context **ret_inode_ac)
2449 {
2450         int ret;
2451         u64 di_blkno;
2452         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2453         struct inode *orphan_dir = NULL;
2454         struct buffer_head *orphan_dir_bh = NULL;
2455         struct ocfs2_alloc_context *inode_ac = NULL;
2456
2457         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2458         if (ret < 0) {
2459                 mlog_errno(ret);
2460                 return ret;
2461         }
2462
2463         /* reserve an inode spot */
2464         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2465         if (ret < 0) {
2466                 if (ret != -ENOSPC)
2467                         mlog_errno(ret);
2468                 goto out;
2469         }
2470
2471         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2472                                        &di_blkno);
2473         if (ret) {
2474                 mlog_errno(ret);
2475                 goto out;
2476         }
2477
2478         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2479                                          di_blkno, orphan_name, orphan_insert,
2480                                          false);
2481         if (ret < 0) {
2482                 mlog_errno(ret);
2483                 goto out;
2484         }
2485
2486 out:
2487         if (ret == 0) {
2488                 *ret_orphan_dir = orphan_dir;
2489                 *ret_di_blkno = di_blkno;
2490                 *ret_inode_ac = inode_ac;
2491                 /*
2492                  * orphan_name and orphan_insert are already up to
2493                  * date via prepare_orphan_dir
2494                  */
2495         } else {
2496                 /* Unroll reserve_new_inode* */
2497                 if (inode_ac)
2498                         ocfs2_free_alloc_context(inode_ac);
2499
2500                 /* Unroll orphan dir locking */
2501                 mutex_unlock(&orphan_dir->i_mutex);
2502                 ocfs2_inode_unlock(orphan_dir, 1);
2503                 iput(orphan_dir);
2504         }
2505
2506         brelse(orphan_dir_bh);
2507
2508         return ret;
2509 }
2510
2511 int ocfs2_create_inode_in_orphan(struct inode *dir,
2512                                  int mode,
2513                                  struct inode **new_inode)
2514 {
2515         int status, did_quota_inode = 0;
2516         struct inode *inode = NULL;
2517         struct inode *orphan_dir = NULL;
2518         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2519         struct ocfs2_dinode *di = NULL;
2520         handle_t *handle = NULL;
2521         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2522         struct buffer_head *parent_di_bh = NULL;
2523         struct buffer_head *new_di_bh = NULL;
2524         struct ocfs2_alloc_context *inode_ac = NULL;
2525         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2526         u64 uninitialized_var(di_blkno), suballoc_loc;
2527         u16 suballoc_bit;
2528
2529         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2530         if (status < 0) {
2531                 if (status != -ENOENT)
2532                         mlog_errno(status);
2533                 return status;
2534         }
2535
2536         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2537                                               orphan_name, &orphan_dir,
2538                                               &di_blkno, &orphan_insert, &inode_ac);
2539         if (status < 0) {
2540                 if (status != -ENOSPC)
2541                         mlog_errno(status);
2542                 goto leave;
2543         }
2544
2545         inode = ocfs2_get_init_inode(dir, mode);
2546         if (IS_ERR(inode)) {
2547                 status = PTR_ERR(inode);
2548                 inode = NULL;
2549                 mlog_errno(status);
2550                 goto leave;
2551         }
2552
2553         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2554         if (IS_ERR(handle)) {
2555                 status = PTR_ERR(handle);
2556                 handle = NULL;
2557                 mlog_errno(status);
2558                 goto leave;
2559         }
2560
2561         status = dquot_alloc_inode(inode);
2562         if (status)
2563                 goto leave;
2564         did_quota_inode = 1;
2565
2566         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2567                                               &suballoc_loc,
2568                                               &suballoc_bit, di_blkno);
2569         if (status < 0) {
2570                 mlog_errno(status);
2571                 goto leave;
2572         }
2573
2574         clear_nlink(inode);
2575         /* do the real work now. */
2576         status = __ocfs2_mknod_locked(dir, inode,
2577                                       0, &new_di_bh, parent_di_bh, handle,
2578                                       inode_ac, di_blkno, suballoc_loc,
2579                                       suballoc_bit);
2580         if (status < 0) {
2581                 mlog_errno(status);
2582                 goto leave;
2583         }
2584
2585         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2586         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2587                                   &orphan_insert, orphan_dir, false);
2588         if (status < 0) {
2589                 mlog_errno(status);
2590                 goto leave;
2591         }
2592
2593         /* get open lock so that only nodes can't remove it from orphan dir. */
2594         status = ocfs2_open_lock(inode);
2595         if (status < 0)
2596                 mlog_errno(status);
2597
2598         insert_inode_hash(inode);
2599 leave:
2600         if (status < 0 && did_quota_inode)
2601                 dquot_free_inode(inode);
2602         if (handle)
2603                 ocfs2_commit_trans(osb, handle);
2604
2605         if (orphan_dir) {
2606                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2607                 ocfs2_inode_unlock(orphan_dir, 1);
2608                 mutex_unlock(&orphan_dir->i_mutex);
2609                 iput(orphan_dir);
2610         }
2611
2612         if ((status < 0) && inode) {
2613                 clear_nlink(inode);
2614                 iput(inode);
2615         }
2616
2617         if (inode_ac)
2618                 ocfs2_free_alloc_context(inode_ac);
2619
2620         brelse(new_di_bh);
2621
2622         if (!status)
2623                 *new_inode = inode;
2624
2625         ocfs2_free_dir_lookup_result(&orphan_insert);
2626
2627         ocfs2_inode_unlock(dir, 1);
2628         brelse(parent_di_bh);
2629         return status;
2630 }
2631
2632 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2633         struct inode *inode)
2634 {
2635         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2636         struct inode *orphan_dir_inode = NULL;
2637         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2638         struct buffer_head *di_bh = NULL;
2639         int status = 0;
2640         handle_t *handle = NULL;
2641         struct ocfs2_dinode *di = NULL;
2642
2643         status = ocfs2_inode_lock(inode, &di_bh, 1);
2644         if (status < 0) {
2645                 mlog_errno(status);
2646                 goto bail;
2647         }
2648
2649         di = (struct ocfs2_dinode *) di_bh->b_data;
2650         /*
2651          * Another append dio crashed?
2652          * If so, manually recover it first.
2653          */
2654         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2655                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2656                 if (status < 0) {
2657                         if (status != -ENOSPC)
2658                                 mlog_errno(status);
2659                         goto bail_unlock_inode;
2660                 }
2661
2662                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2663                 if (status < 0) {
2664                         mlog_errno(status);
2665                         goto bail_unlock_inode;
2666                 }
2667         }
2668
2669         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2670                         OCFS2_I(inode)->ip_blkno,
2671                         orphan_name,
2672                         &orphan_insert,
2673                         true);
2674         if (status < 0) {
2675                 mlog_errno(status);
2676                 goto bail_unlock_inode;
2677         }
2678
2679         handle = ocfs2_start_trans(osb,
2680                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2681         if (IS_ERR(handle)) {
2682                 status = PTR_ERR(handle);
2683                 goto bail_unlock_orphan;
2684         }
2685
2686         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2687                         &orphan_insert, orphan_dir_inode, true);
2688         if (status)
2689                 mlog_errno(status);
2690
2691         ocfs2_commit_trans(osb, handle);
2692
2693 bail_unlock_orphan:
2694         ocfs2_inode_unlock(orphan_dir_inode, 1);
2695         mutex_unlock(&orphan_dir_inode->i_mutex);
2696         iput(orphan_dir_inode);
2697
2698         ocfs2_free_dir_lookup_result(&orphan_insert);
2699
2700 bail_unlock_inode:
2701         ocfs2_inode_unlock(inode, 1);
2702         brelse(di_bh);
2703
2704 bail:
2705         return status;
2706 }
2707
2708 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2709                 struct inode *inode, struct buffer_head *di_bh,
2710                 int update_isize, loff_t end)
2711 {
2712         struct inode *orphan_dir_inode = NULL;
2713         struct buffer_head *orphan_dir_bh = NULL;
2714         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2715         handle_t *handle = NULL;
2716         int status = 0;
2717
2718         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2719                         ORPHAN_DIR_SYSTEM_INODE,
2720                         le16_to_cpu(di->i_dio_orphaned_slot));
2721         if (!orphan_dir_inode) {
2722                 status = -ENOENT;
2723                 mlog_errno(status);
2724                 goto bail;
2725         }
2726
2727         mutex_lock(&orphan_dir_inode->i_mutex);
2728         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2729         if (status < 0) {
2730                 mutex_unlock(&orphan_dir_inode->i_mutex);
2731                 iput(orphan_dir_inode);
2732                 mlog_errno(status);
2733                 goto bail;
2734         }
2735
2736         handle = ocfs2_start_trans(osb,
2737                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2738         if (IS_ERR(handle)) {
2739                 status = PTR_ERR(handle);
2740                 goto bail_unlock_orphan;
2741         }
2742
2743         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2744
2745         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2746                                 inode, orphan_dir_bh, true);
2747         if (status < 0) {
2748                 mlog_errno(status);
2749                 goto bail_commit;
2750         }
2751
2752         status = ocfs2_journal_access_di(handle,
2753                         INODE_CACHE(inode),
2754                         di_bh,
2755                         OCFS2_JOURNAL_ACCESS_WRITE);
2756         if (status < 0) {
2757                 mlog_errno(status);
2758                 goto bail_commit;
2759         }
2760
2761         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2762         di->i_dio_orphaned_slot = 0;
2763
2764         if (update_isize) {
2765                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2766                 if (status)
2767                         mlog_errno(status);
2768         } else
2769                 ocfs2_journal_dirty(handle, di_bh);
2770
2771 bail_commit:
2772         ocfs2_commit_trans(osb, handle);
2773
2774 bail_unlock_orphan:
2775         ocfs2_inode_unlock(orphan_dir_inode, 1);
2776         mutex_unlock(&orphan_dir_inode->i_mutex);
2777         brelse(orphan_dir_bh);
2778         iput(orphan_dir_inode);
2779
2780 bail:
2781         return status;
2782 }
2783
2784 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2785                                    struct inode *inode,
2786                                    struct dentry *dentry)
2787 {
2788         int status = 0;
2789         struct buffer_head *parent_di_bh = NULL;
2790         handle_t *handle = NULL;
2791         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2792         struct ocfs2_dinode *dir_di, *di;
2793         struct inode *orphan_dir_inode = NULL;
2794         struct buffer_head *orphan_dir_bh = NULL;
2795         struct buffer_head *di_bh = NULL;
2796         struct ocfs2_dir_lookup_result lookup = { NULL, };
2797
2798         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2799                                 dentry->d_name.len, dentry->d_name.name,
2800                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2801                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2802
2803         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2804         if (status < 0) {
2805                 if (status != -ENOENT)
2806                         mlog_errno(status);
2807                 return status;
2808         }
2809
2810         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2811         if (!dir_di->i_links_count) {
2812                 /* can't make a file in a deleted directory. */
2813                 status = -ENOENT;
2814                 goto leave;
2815         }
2816
2817         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2818                                            dentry->d_name.len);
2819         if (status)
2820                 goto leave;
2821
2822         /* get a spot inside the dir. */
2823         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2824                                               dentry->d_name.name,
2825                                               dentry->d_name.len, &lookup);
2826         if (status < 0) {
2827                 mlog_errno(status);
2828                 goto leave;
2829         }
2830
2831         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2832                                                        ORPHAN_DIR_SYSTEM_INODE,
2833                                                        osb->slot_num);
2834         if (!orphan_dir_inode) {
2835                 status = -ENOENT;
2836                 mlog_errno(status);
2837                 goto leave;
2838         }
2839
2840         mutex_lock(&orphan_dir_inode->i_mutex);
2841
2842         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2843         if (status < 0) {
2844                 mlog_errno(status);
2845                 mutex_unlock(&orphan_dir_inode->i_mutex);
2846                 iput(orphan_dir_inode);
2847                 goto leave;
2848         }
2849
2850         status = ocfs2_read_inode_block(inode, &di_bh);
2851         if (status < 0) {
2852                 mlog_errno(status);
2853                 goto orphan_unlock;
2854         }
2855
2856         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2857         if (IS_ERR(handle)) {
2858                 status = PTR_ERR(handle);
2859                 handle = NULL;
2860                 mlog_errno(status);
2861                 goto orphan_unlock;
2862         }
2863
2864         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2865                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2866         if (status < 0) {
2867                 mlog_errno(status);
2868                 goto out_commit;
2869         }
2870
2871         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2872                                   orphan_dir_bh, false);
2873         if (status < 0) {
2874                 mlog_errno(status);
2875                 goto out_commit;
2876         }
2877
2878         di = (struct ocfs2_dinode *)di_bh->b_data;
2879         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2880         di->i_orphaned_slot = 0;
2881         set_nlink(inode, 1);
2882         ocfs2_set_links_count(di, inode->i_nlink);
2883         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2884         ocfs2_journal_dirty(handle, di_bh);
2885
2886         status = ocfs2_add_entry(handle, dentry, inode,
2887                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2888                                  &lookup);
2889         if (status < 0) {
2890                 mlog_errno(status);
2891                 goto out_commit;
2892         }
2893
2894         status = ocfs2_dentry_attach_lock(dentry, inode,
2895                                           OCFS2_I(dir)->ip_blkno);
2896         if (status) {
2897                 mlog_errno(status);
2898                 goto out_commit;
2899         }
2900
2901         d_instantiate(dentry, inode);
2902         status = 0;
2903 out_commit:
2904         ocfs2_commit_trans(osb, handle);
2905 orphan_unlock:
2906         ocfs2_inode_unlock(orphan_dir_inode, 1);
2907         mutex_unlock(&orphan_dir_inode->i_mutex);
2908         iput(orphan_dir_inode);
2909 leave:
2910
2911         ocfs2_inode_unlock(dir, 1);
2912
2913         brelse(di_bh);
2914         brelse(parent_di_bh);
2915         brelse(orphan_dir_bh);
2916
2917         ocfs2_free_dir_lookup_result(&lookup);
2918
2919         if (status)
2920                 mlog_errno(status);
2921
2922         return status;
2923 }
2924
2925 const struct inode_operations ocfs2_dir_iops = {
2926         .create         = ocfs2_create,
2927         .lookup         = ocfs2_lookup,
2928         .link           = ocfs2_link,
2929         .unlink         = ocfs2_unlink,
2930         .rmdir          = ocfs2_unlink,
2931         .symlink        = ocfs2_symlink,
2932         .mkdir          = ocfs2_mkdir,
2933         .mknod          = ocfs2_mknod,
2934         .rename         = ocfs2_rename,
2935         .setattr        = ocfs2_setattr,
2936         .getattr        = ocfs2_getattr,
2937         .permission     = ocfs2_permission,
2938         .setxattr       = generic_setxattr,
2939         .getxattr       = generic_getxattr,
2940         .listxattr      = ocfs2_listxattr,
2941         .removexattr    = generic_removexattr,
2942         .fiemap         = ocfs2_fiemap,
2943         .get_acl        = ocfs2_iop_get_acl,
2944         .set_acl        = ocfs2_iop_set_acl,
2945 };